瞄准未来高技术战争,选择地面战车为研究对象,系统地研究了红外图像自动目标识别的若干关键技术。首先比较了三种建立红外图像数据集方法的性能。接着分析了传统红外图像分割技术的缺陷,提出了一种针对地面背景中战车红外图像的分割技术,该方法有效地克服了过分割和欠分割的问题。然后讨论了红外目标的七大类特征,提出了一种确定最优特征向量的方法。最后以支持向量机作为分类器,讨论了核函数、核参数以及惩罚因子对识别率的影响,并运用网格搜索法确定最优的参数组合。

Focusing on the high technology war in the future, the key techniques for infrared image automatic target recognition on the ground chariots are studied by the numbers. Firstly, the performance of three methods that setting up infrared image data sets is compared. Secondly, limitation of the traditional infrared image segmentation methods is analysed. A new segmentation method for the infrared image of chariots on the ground is put forward. This method overcomes the problems of over-segmentation and less-segmentation. Thirdly, seven kinds of features of infrared targets are discussed. A new method that finds out the best feature vector is advanced. At last, using support vector machine as the classification machine, the influence of kernel functions, kernel function parameters and penalty factor on the recognition rate is discussed. The optimal parameters combination is fixed by grid-search algorithm.